Physical Spark — build the engine

What we need

Four openings, and none of them are "help out sometime"

These are the specific things blocking the next season. If one of them is obviously yours, that is the whole application — tell us which and why.

01 · ROBOTICS

Mission designer

Turn a robot task into something that can be graded from a video. You have run LeRobot or an SO-101-class arm and know where beginners actually stall.

02 · DATA

Data pipeline

Every run is an episode. Someone has to make them consistent, labelled and worth paying for — formats, dedup, quality gates, the boring parts that decide whether a lab buys.

03 · PRODUCT

Full-stack

Upload, peer review, leaderboard, seasons. Small stack, no framework religion — the site you are reading is vanilla HTML and a tiny Python server, and it deploys on a git push.

04 · COMMUNITY

First-season host

Run the room. Recruit the first cohort, keep reviews moving, and notice why people stop — completion rate is the number the whole company is judged on.

Who fits

You would rather ship a rough thing than plan a clean one

You finish. The product is a completion rate. People who start things are easy to find.
You are new to robotics, or you are not — both work. Two of the four roles above have nothing to do with kinematics. What does not work is waiting until you feel qualified.
You are comfortable in public. Everything here is built in the open, mistakes included. If a wrong commit in your name would keep you up, this will be uncomfortable.
You can read English or Korean. The team works across both. The product is English.

Straight about where this is: pre-funding, pre-revenue, no salary yet. There is a real business model (three layers, one loop →) and a team that has done the underlying work before, but today the honest offer is equity-shaped, not cash-shaped. Everyone keeps their day job for now. If you need a salary this quarter, take the salary — and come back when we can pay you properly.

What you get

Early enough that your taste still shows

Direction, not tickets. Nothing here is settled. Which missions exist, what a good demonstration looks like, how a season is scored — these are being decided now, and whoever is in the room decides them.

A public record. Your commits, your missions, your reviews. The same artefact the platform gives its users, you build for yourself first.

Physical AI, hands on the hardware. Not a course about robots — an arm on a desk, a camera pointed at it, and data coming out the other side.

A short path to being wrong. Seasons are weeks, not quarters. You will find out whether an idea worked while you still remember having it.

How to start

Open an issue. That is the whole process

STEP 1

Say what you would take on

Open a GitHub issue naming one of the four roles, or something we have not thought of. A paragraph is enough. No CV needed — though a link to something you finished helps more than one.

STEP 2

Do one small real thing

Pick anything labelled good first issue, or fix something that annoyed you on this site. Working together for an afternoon tells us both more than an interview would.

STEP 3

Decide together

If it works, we talk about scope and equity properly. If it does not, you keep the commit and we both saved a month.

Come build
the engine.